Created attachment 648288 [details] [diff] [review]
Kyle mentioned this to me the other day, but I just ran into it myself. Relatively recent regression, but don't have a range.
(1) Build b2g for desktop (https://wiki.mozilla.org/Gaia/Hacking#Building_B2G)
(2) Launch with MOZ_USE_OMTC=1 for omtc
We abort pretty soon after startup in a debug build.
The attached patch makes the crash stop, but it doesn't shed any light on the underlying problem.
Created attachment 648478 [details]
Can someone who can reproduce compile this with "gcc -lGL glxinfo.c" and attach the output, please?
I'm suspicious of bug 707722, but that's not really "the other day".
Does this happen only with MOZ_USE_OMTC=1?
Does that turn on GL layers, or is that orthogonal?
Created attachment 648481 [details]
Previous attachment was wrong file.
Bug 707722 was based on http://www.opengl.org/sdk/docs/man/xhtml/glXGetFBConfigAttrib.xml
Number of bits per color buffer.
If the frame buffer configuration supports RGBA contexts, then
GLX_BUFFER_SIZE is the sum of GLX_RED_SIZE, GLX_GREEN_SIZE,
GLX_BLUE_SIZE, and GLX_ALPHA_SIZE.
If the frame buffer configuration supports only color index contexts,
GLX_BUFFER_SIZE is the size of the
Apparently NVIDIA fbconfigs may still have quite a different interpretation.
i.e. GLX_BUFFER_SIZE always 32, even with no alpha bits.
e.g. attachment 519303 [details]
Created attachment 649177 [details] [diff] [review]
check alpha instead of buffer size against render format
The assumption that alpha and color bits filled up the render format depth wasn't correct anyway.
We don't currently run reftests with GLX layers but they do surprisingly well.
Just some border-radius tests that also fail on other platforms, and a different "Failed to create pixmap!" in scrolling/image-1.html which is kinda expected when trying to GLXCreatePixmap from a gfxImageSurface.
*** Bug 779852 has been marked as a duplicate of this bug. ***